Class DistributedProxySampler

datasets/DistributedProxySampler.py:8–45

Sampler that restricts data loading to a subset of input sampler indices. It is especially useful in conjunction with :class:`torch.nn.parallel.DistributedDataParallel`. In such case, each process can pass a DistributedSampler instance as a DataLoader sampler, and load a subset of t

class DistributedProxySampler(DistributedSampler):
    """Sampler that restricts data loading to a subset of input sampler indices.

    It is especially useful in conjunction with
    :class:`torch.nn.parallel.DistributedDataParallel`. In such case, each
    process can pass a DistributedSampler instance as a DataLoader sampler,
    and load a subset of the original dataset that is exclusive to it.

    .. note::
        Input sampler is assumed to be of constant size.

    Arguments:
        sampler: Input data sampler.
        num_replicas (optional): Number of processes participating in
            distributed training.
        rank (optional): Rank of the current process within num_replicas.
    """

    def __init__(self, sampler, num_replicas=None, rank=None):        
        super(DistributedProxySampler, self).__init__(sampler, num_replicas=num_replicas, rank=rank, shuffle=False)
        self.sampler = sampler

    def __iter__(self):
        # deterministically shuffle based on epoch
        torch.manual_seed(self.epoch)
        indices = list(self.sampler)

        # add extra samples to make it evenly divisible
        indices += indices[:(self.total_size - len(indices))]
        if len(indices) != self.total_size:
            raise RuntimeError("{} vs {}".format(len(indices), self.total_size))

        # subsample
        indices = indices[self.rank:self.total_size:self.num_replicas]
        if len(indices) != self.num_samples:
            raise RuntimeError("{} vs {}".format(len(indices), self.num_samples))

        return iter(indices)